Browse Prior Art Database

Length-Normalized Two-Level DP Matching Method for Continuous Speech Recognition

IP.com Disclosure Number: IPCOM000044174D
Original Publication Date: 1984-Nov-01
Included in the Prior Art Database: 2005-Feb-05
Document File: 2 page(s) / 78K

Publishing Venue

IBM

Related People

Matsuda, Y: AUTHOR

Abstract

This technique provides connected-words speech recognition by two-level DP matching with a better recognition score and reduction of computation time. Two-level DP matching has already been disclosed as a tool for connected-words speech recognition [*]. Table 1 shows recognition results for 35 x 4-digit words by three male speakers using a conventional two-level DP matching method [*]. The following results are obtained from this table. (1) Input (Inp) utterance speed for one digit is faster than reference (Ref) utterance speed. (2) Recognition results for the person (e.g., speaker N) who speaks faster in input utterance become worse.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 60% of the total text.

Page 1 of 2

Length-Normalized Two-Level DP Matching Method for Continuous Speech Recognition

This technique provides connected-words speech recognition by two-level DP matching with a better recognition score and reduction of computation time. Two- level DP matching has already been disclosed as a tool for connected-words speech recognition [*]. Table 1 shows recognition results for 35 x 4-digit words by three male speakers using a conventional two-level DP matching method [*]. The following results are obtained from this table. (1) Input (Inp) utterance speed for one digit is faster than reference (Ref) utterance speed. (2) Recognition results for the person (e.g., speaker N) who speaks faster in input utterance become worse. When the reference speed for each digit is the same as the input utterance speed, the average input utterance length of connected 4-digit words (Inp) will become four times the average reference utterance length (Ref). The new technique or Length Normalized two-level DP Matching method (LN2DP) was developed in order to obtain a higher recognition result for the person who speaks relatively fast. LN2DP may be described as follows: (1) Assumption Each input pattern is composed of several reference patterns and the number of references for one input(n) is known, such as connected digit recognition with a known number of digits. (2) Procedures (i) Each utterance length of reference patterns is normalized by Ref. (ii) The utterance length of the input patte...